Skip to main content
luminarforge.
  • Real people & productsWritingTemplatesIllustrationsPublishingPlanning
  • For agencies
  • For one business
  • Examples
  • Pricing
Log in Try it free
The pitch desk — AI suggestions

The blank page, abolished.

The desk deals pitch slips built from the client’s own brand brain — topic, platform, format and a suggested day, and never more than two on the same platform. Take one and it opens a new post with the topic and platform filled in. Spike a theme three times and it sinks to the bottom of the pile for a fortnight.

Start free — 5,000 credits How the desk pitches ↓

Suggestions on every plan · credit-metered — opening the desk can regenerate the pile, which is charged

Ideas argued from their knowledge base — not a trending-topics feed.

Fig. 1 — a dealt pile · Alma Skin Illustrative slips · four real fields and a movable day
Taken → new post
Pitch 01 · LinkedIn essayTheme: barrier science
Why “more moisturiser” is the wrong answer

Suggested: Tuesday

Pitch 02 · Instagram carouselTheme: routine building
The 3-step evening reset, illustrated

Suggested: Thursday

Pitch 03 · Blog long-formTheme: ingredient literacy
Ceramides, explained like a label-reader

Suggested: Monday

Spiked ×3 — sinks for a fortnight
Pitch — · LinkedInTheme: founder diary
A week in the founder’s skincare routine

Spiked: the one signal it learns from

The desk’s methods — where pitches come from

Argued from the brand brain, filtered like an editor.

Pitches aren’t random and they aren’t trending-topic scrapes — every slip traces back to the client’s own knowledge base, and the desk remembers what you spike.

Method 01 — Themes

The theme map

The knowledge base is distilled into weighted core themes per client — pitches draw from the map, so a skincare brand gets barrier science, not generic “5 productivity hacks”.

Weighted per client · refreshed with the KB
Method 02 — Empty days

The dashboard counts your empty days

The dashboard counts the days your calendar has left empty and prompts you against them. The pile itself is not ordered by date — each pitch carries a suggested day, which you can move.

Empty-day prompts on the dashboard · a suggested day on every slip
Method 03 — Variety rules

Two per platform, and nothing thrown away

No more than two pitches on the same platform reach you at once, so one channel cannot take the pile over. Nothing is merged and nothing is deleted — the pile is ordered, not filtered, and a low-scoring pitch can still be dealt when there is little else.

Max two per platform · nothing merged, nothing dropped
Method 04 — Dismissal memory

It learns what you spike

Dismiss a pitch and it’s remembered. Spike the same theme three times and it sinks to the bottom of the pile — still there if the pile is thin, and back at full weight after a fortnight. It’s the one learning behaviour we claim, because it’s the one that exists.

Theme demoted at 3+ dismissals · that penalty ages out after 14 days
After the presses run — the returns loop

What the engagement data will teach — the thresholds, printed in advance.

Published posts report their engagement back. Three instruments are written to read it, each with a printed threshold — because statistics on four posts would be astrology. None of the three is switched on yet: the thresholds below are the gates they will use, not gates you are approaching.

Instrument 01 — Prediction

Written to estimate a rate, not a reach

Written to estimate a post’s engagement rate from your own history — hook style, length, day, quality score. Not wired to anything you can open, and the history it reads is not being recorded yet.

The gate, when it opens · ≈10 published posts
Instrument 02 — Calibration

Written to learn your audience

Correlation maths that would check which quality dimensions actually predict engagement for this client. No score is re-weighted today — the six dimension weights are fixed, and nothing in the product can start one.

The gate, when it opens · 20+ samples at 0.7 confidence
Instrument 03 — Fatigue

Written to spot a tired topic

Watches the trend across repeated keywords so a real decline flags the topic. The badge is not shown anywhere in the product.

The gate, when it opens · 3+ posts per keyword in 90 days

Straight answer: none of these three is running yet. The code and the thresholds are real and printed above so you can hold us to them; what they don’t do is steer which topics the pitch desk deals, or change your score. Two desks, side by side, and only one of them open. When any of it switches on, the changelog says so first.

The scoring gate itself lives on the AI Pipeline page
Said plainly — how we keep this honest
On “it learns”

Today the pitch desk learns from one signal: your dismissals. It does not yet learn from what you use, save, or edit — plenty of tools imply otherwise about themselves; we’d rather under-claim and ship the rest later.

On predictions

There are no predictions yet. The code is written and its gate is printed above, but nothing in the product calls it and the history it would read is not being recorded. When it runs it will give an engagement-rate estimate from your own posts — never a promised reach number, and never a crystal ball.

The pile is dealt

Take tomorrow’s pitch.

Start free with 5,000 credits — paste a client’s URL and the desk starts pitching from their world the same day. Roughly a fortnight of real production, on us.

Start with 5,000 free credits See the rate card

No card · No trial countdown · No per-client fees

luminarforge.

Product

  • Real people & products
  • Writing
  • Templates
  • Illustrations
  • Publishing
  • Planning

Who it's for

  • For agencies
  • For one business
  • Examples
  • Compare
  • vs Canva
  • Pricing

Company

  • Trust & security
  • Changelog
  • Contact
  • Log in

Legal

  • Terms
  • Privacy
  • Cookies
  • Refunds

Publishes to LinkedIn profiles and company pages. Instagram, Facebook and X: copy-and-paste.

© 2026 Luminar Works